Central nervous system (CNS) inflammation involves the generation of inducible cytokines such as interferons (IFNs) and alterations in brain activity, yet the interplay of both is not well understood. Here, we show that in vivo elevation of IFNs by viral brain infection reduced hyperpolarization-activated currents (Ih) in cortical pyramidal neurons. In rodent brain slices directly exposed to type I IFNs, the hyperpolarization-activated cyclic nucleotide (HCN)-gated channel subunit HCN1 was specifically affected. The effect required an intact type I receptor (IFNAR) signaling cascade. Consistent with Ih inhibition, IFNs hyperpolarized the resting membrane potential, shifted the resonance frequency, and increased the membrane impedance. In vivo application of IFN-β to the rat and to the mouse cerebral cortex reduced the power of higher frequencies in the cortical electroencephalographic activity only in the presence of HCN1. In summary, these findings identify HCN1 channels as a novel neural target for type I IFNs providing the possibility to tune neural responses during the complex event of a CNS inflammation.
The cytokines interferons (IFN)-α and -β belong to type I IFNs and share receptor and signaling pathways. They are key elements in the mammalian first-line innate immune response because they activate immune cells, stimulate surface molecules, and regulate the differentiation of monocytes (Pestka 2007). In the context of central nervous system (CNS) inflammation, IFN-β is increased in myeloid cells (Prinz et al. 2008) and locally produced by neurons (Delhaye et al. 2006). In turn, all cell types of the CNS respond to type I IFNs (Delhaye et al. 2006; Paul et al. 2007; Detje et al. 2009). Besides these established immunological and antiviral effects, evidence for a neuromodulatory potential of IFNs is growing. For example, during medical treatment, IFNs can lead to various behavioral changes, for example, fatigue, cognitive dysfunction, depressed mood, condensed as sickness behavior (Dantzer et al. 2008). On the cellular level, type I IFNs enhance neuronal excitability in neurons of the cortex, the hippocampus, and the amygdala (Dafny et al. 1996). However, the mechanisms by which type I IFNs impact on neuronal excitability are poorly understood.
In our previous in vitro study, application of IFN-β lead to an increased firing rate and input resistance in cortical neurons (Hadjilambreva et al. 2005). Indirect evidence suggested that the hyperpolarization-activated nonselective cation current (Ih) may serve as a possible molecular target of IFNs: IFN-β decreased neuronal resting conductance and blocking Ih prevented the increase in input resistance after IFN-β application (Hadjilambreva et al. 2005). In addition, inflammation altered Ih in myenteric neurons (Linden et al. 2003). However, interactions between ion channels and IFNs have not been investigated in detail.
Ih is mediated by HCN channels, which derive from four genes (HCN1–4) found throughout the brain (Santoro et al. 1998). HCN channel subunits generate channels with distinct biophysical properties by assembling to homo- or heterotetrameric complexes (Frere et al. 2004). In neocortical pyramidal neurons, Ih is mainly mediated by HCN1 and HCN2. HCN1 is highly expressed in distal dendrites where it regulates action potential firing patterns (Kole et al. 2006), synaptic integration (Stuart and Spruston 1998; Strauss et al. 2004) and contributes to the subthreshold somato-dendritic voltage attenuation (Zhang 2004) and membrane resonance (Narayanan and Johnston 2008). Modulation of Ih provides a cellular mechanism to alter network oscillations of neural circuits during cognitive processing (Wahl-Schott and Biel 2009). HCN channels are sensitive to a number of intra- and extracellular modulators, which in many cases act by shifting the channel's voltage sensitivity via either cAMP, intracellular protons, phosphatidylinositol-4,5-phosphate or acidic lipids (for review: Wahl-Schott and Biel 2009).
To address whether and, if so, how the neuromodulation by IFNs is mediated via Ih changes, we investigated the effect of type I IFNs on HCN channels of rat neocortical layer 5 neurons. The results show a reduction and deceleration of Ih with consequences for the neuronal frequency behavior. This effect is mediated by the IFN-signaling cascade and specifically governed by a modulation of the fastest activating HCN subunit, HCN1. The data emphasize the role of cytokines in determining the single neuron and network states and present the first direct evidence for a type I IFN action on a neuronal ion channel.
Materials and Methods
Interferons and Signaling Pathway Modulators
For all experiments, Chinese hamster ovary-derived recombinant IFN (rat IFN-α,-β: U-CyTech, Utrecht, The Netherlands; mouse IFN-β: Hycultec GmbH, Beutelsbach, Germany) was used. The lyophilized product was reconstituted in sterile double-distilled water, and small aliquots were stored according to the data sheets. The activation of Janus protein tyrosine kinase (JAK)1 or JAK2 was prevented by the selective blocker substances JAK inhibitor1 and AG-490, respectively (both from Calbiochem, San Diego, CA, United States of America). Blockers were dissolved in dimethyl sulfoxide to 75 or 500 µM, respectively, and stored at −20 °C.
Male Wistar rats (Forschungseinrichtung für experimentelle Medizin (FEM), Berlin, Germany) and mice (C57/B6/J and B6/129-HCN1tm2Kndl/J, Jackson Laboratory) were bred at the local animal facility. For HCN1−/− experiments B6/129-HCN1tm2Kndl/J were crossed with C57/B6/J, heterozygote offspring were further crossed, and genotyping was performed from tail cuts via polymerase chain reaction (PCR) according to the available protocol (Jackson Laboratory). Animals were kept under standard laboratory conditions, and all procedures were performed in agreement with the European Communities Council Directive of 24 November 1986 (86/609/EEC).
Anesthetized (ketamine, 100 mg kg−1 intraperitoneal (i.p.); DeltaSelect and xylazine 20 mg kg−1 i.p.; Bayer Health Care, Berlin, Germany) 21-day-old male C57/B6 mice were intrathecally (L2 or L3) injected (Hoffmann et al. 2007) with 20 µL of phosphate-buffered saline (PBS; 3 mice) or 25 µL virus stock (9 mice) containing 106 PFU Theiler's murine encephalomyelitis virus (TMEV) GDVII (Delhaye et al. 2006). Postoperatively, adequate waking and the absence of paresis were verified. Experimental procedures were reviewed by institutional and state authorities (G0175/07).
Slice Preparation and Culture of Cortical Neurons and HEK293 Cells
For a detailed description see Supplementary Methods.
Individual slices, HEK293 cells, or primary cultures were transferred to a submerged recording chamber. Cortical neurons and HEK293 cells were visualized with either an Axioskop 2 FS Zeiss or an Axiovert S100 (both from Carl Zeiss MicroImaging GmbH, Göttingen, Germany), and whole-cell patch-clamp experiments were carried out at room temperature (RT; 22–24 °C). For the visualization of cortical pyramidal neurons, infrared differential interference contrast video microscopy was used. These experiments were accomplished at 32–34 °C. Patch pipettes were pulled to a final resistance of 3–5 MΩ. In somatic whole-cell voltage clamp, a maximal series resistance of 15 MΩ with changes < 20% during recordings was tolerated. A fast (pipette) capacitive transient (τ < 1.5 μs, 6–13pF) was compensated. The pipette solution contained (in mM): 120 KMeSO4 (ICN Biomedicals, Eschwege, Germany), 20 KCl (Merck), 14 Na-phospocreatine, 0.5 ethylene glycol tetraacetic acid, 4 NaCl, 10 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, 4 Mg2+-ATP, 0.3 Tris3-GTP with or without 0.1 cAMP (all from Sigma-Aldrich, pH 7.25, 288mOsm). A liquid junction potential of −10 mV has not been corrected for. Voltage-clamp recordings of pharmacologically isolated Ih were obtained by blocking IK(IR) with 200–400 µM Ba2+ added to a modified artificial cerebrospinal fluid: 10 mM K+, MgCl2 replacing MgSO4 (all from Merck), and NaH2PO4 omitted. Sodium currents were suppressed with 1 µM tetrodotoxin (Tocris, Bristol, United Kingdom), low-threshold Ca2+ currents were blocked with 1 mM Ni2+, glutamate receptors were blocked with 20 µM 6-cyano-7-nitroquinoxaline-2,3-dione (both Sigma-Aldrich), and 25 µM D-(-)-2-amino-5-phosphonopentanoic acid (Tocris), GABAA receptors were blocked with 10 µM bicuculline (Tocris), K+ currents (fast-inactivating A-type, IA and delayed rectifier type, IK(DR)) were blocked with 5 mM 4-aminopyridine and 10 mM tetraethylammonium (both from Sigma-Aldrich). Data from all patch-clamp recordings were collected with an EPC-10 (HEKA Elektronik GmbH, Lambrecht, Germany), digitized (10 kHz, after Bessel filtering at 2.5 kHz), and stored using PATCHMASTER software (HEKA). The electrical resonance of the neurons was analyzed with the impedance (Z) amplitude profile (ZAP) method (Narayanan and Johnston 2008). For details see Supplementary Experimental Procedures.
For simulations of membrane resonance, we implemented an established model (Hutcheon et al. 1996) in LabVIEW 8.6 (National Instruments Inc., TX, United States of America) and Scilab 5.0.1 (Scilab Consortium, INRIA, ENPC and Contributors, 1989–2008) and utilized a NEURON model (version 7.1; see Supplementary Experimental Procedures) using a reconstructed morphology of a layer 5 pyramidal neuron (Stuart and Spruston 1998, Fig. 1A therein) with properties described in detail in Supplementary Experimental Procedures. The latter was also applied to test the contribution of the different HCN subtypes.
Quantitative Real-Time PCR
For details see Supplementary Methods.
Sagittal brain sections (30 µm) were prepared as described (Kole et al. 2007). The endogenous peroxidase was quenched by 0.3% hydrogen peroxide diluted in PBS for 30 min at RT. After washing, brain sections were blocked with 10% fetal bovine serum (FBS) in PBS overnight at 4 °C. Sections were incubated with anti-IFNAR1 (Abcam, Cambridge, United Kingdom, diluted 1:50 in blocking solution) at 4 °C for 48 h. After washing in PBS, sections were incubated with a biotinylated secondary anti-rabbit antibody (Invitrogen, Darmstadt, Germany) diluted 1:400 in PBS overnight at 4 °C and then in avidin–biotin peroxidase complex reagent (Vectastain ABC Kit, Vector Labs, Burlingame, CA, United States of America) for 2 h at RT. Immunoreaction was visualized with 3,3′-diaminobenzidine as a chromogen. Cultured primary neurons were fixed with 4% paraformaldehyde and 15% sucrose for 20 min at RT. Cells were treated with 20% FBS and incubated with anti-IFNAR1 antibody diluted 1:100 in 5% FBS. After washing with PBS, cells were incubated with a secondary Alexa 488-labeled goat anti-rabbit antibody (Invitrogen; diluted 1:1500). Afterwards, cells were treated with 0.1% Triton X-100 for 3 min and again blocked with 20% FBS. Anti-mitogen-activated protein 2 (MAP2; Sigma-Aldrich) was incubated in a dilution of 1:1500 in 5% FBS. The cells were washed thrice, and Alexa 568-labeled goat anti-mouse antibody (Invitrogen) was added in a dilution of 1:1500. All incubations were performed overnight at 4 °C. Cells were imaged using an upright confocal microscope (DM250, Leica Microsystems CMS GmbH, Mannheim, Germany). For further details and cultures of glial cells see Supplementary Experimental Procedures.
IFN-β–treated primary neurons and respective controls were lysed in ice-cold buffer (150 mM NaCl, 1% NP-40, 25 mM MgCl2, 10% glycerin, 50 mM Tris, pH 7.4) containing complete protease inhibitor cocktail and phosphatase inhibitors (both Roche, Mannheim, Germany). Protein extracts were separated on SDS–PAGE and electroblotted to a nitrocellulose membrane. Membranes were blocked for 3 h at RT with 5% BSA and incubated with the following antibodies: Anti-p38 MAP kinase 1:1000; anti-phospho-p38 MAP kinase (Thr180/Tyr182) 1:250, anti-phospho-Tyk2 (Tyr1054/1055) 1:1000 (all Cell Signaling, Danvers, MA, United States of America), or anti-Tyk2 1:200 (Santa Cruz Biotechnology, Santa Cruz, CA, United States of America) overnight at 4 °C. Blots were incubated with horseradish peroxidase-conjugated secondary antibodies (1:5000, GE Healthcare, Chalfont St Giles, United Kingdom) overnight at 4 °C, and immunoreactive bands were visualized with chemoluminescence.
Cortical activity from freely moving Wistar rats was recorded on postnatal days (P)11–12 and from freely moving C57/B6/J and B6/129-HCN1tm2Kndl/J mice on P32–40. For rats: Cortical electrodes were placed and fixed at P10–11 as described previously (Schuchmann et al. 2006) at the following coordinates: 3 mm posterior from the bregma, 2 mm lateral from the midline, with a subdural reference electrode above the cerebellum. Close to the recording electrode, a guiding plastic tube (conical, tip outer diameter 0.9 mm) was implanted above the dura (2 mm posterior from the bregma, 2 mm lateral from the midline; Fig. 8A). After recovery from anesthesia, the pups were returned into their original litter; experiments started earliest 24 h after surgery. The recordings were performed in a specific heated area (32 °C; Supplementary Fig. S8A). The signals were sampled at 0.5–3 kHz using a 12-bit data acquisition board (AD Instruments GmbH, Spechbach, Germany). After a period of exploration (30–60min), the animals started to show increasing silent periods, which were recorded for at least 2 h. Using the implanted guiding plastic tube, we applied rat IFN-β. To reach a large number of neurons and an immediate onset of IFN-β effect, we used a time delayed application technique: The dura was perforated through the implanted guiding tube with a 26-Gauge needle, then we applied 5000 IU IFN-β in 5 µL saline via a 30-Gauge Hamilton syringe into the tubing on top of the perforated dura mater. Using this technique, we avoided any volume effect to the cortex. Subsequently, we recorded again for at least 1h. For the analysis, we only used periods with motorically silent behavior. In control experiments, only saline was applied.
For mice, electroencephalography (EEG) experiments were performed as in rats with slight modifications: Between P30–34 cortical electrodes were placed and fixed 1.5 mm posterior from the bregma and 2 mm lateral from the midline. Saline and IFN-β injections were done in identical animals at least 48 h apart (Fig. 9A) to increase comparability and to reduce animal numbers. Because the mice never were motorically silent, periods showing the least motor artifacts were analyzed.
For all data, statistics were performed depending on the dataset as appropriate in Origin7 (OriginLab, Northhampton, MA, United States of America) or Statview v.457 (Abacus Concepts Inc., CA, United States of America). In detail, for normal distributed datasets (Shapiro-Wilk W-test), we used 2-tailed Student's t-tests. In the case of significant deviations from the normal distribution (P ≤ 0.05) or if the sample size was too small (n ≤ 8) for a reliable test of normal distribution, nonparametric tests were used: Wilcoxon signed-rank test for paired sample sets and Mann–Whitney U-test for unpaired sample sets. Data are presented as mean ± SD.
Viral Infection Inhibits Neocortical Ih
Neurons in mice infected with the neurovirulent GDVII strain of TMEV act as producers of type I IFNs after intrathecal virus application (Delhaye et al. 2006). To test whether in vivo IFN induction by a CNS infection is influencing Ih, we used the TMEV model and performed whole-cell recordings on somatosensory cortex layer 5 neurons at the third or fourth day postinfection (Fig. 1A,B), when animals started to show signs of sickness indicating the elevation of early induced type I IFN levels. TMEV infection reduced the amplitudes of pharmacologically isolated Ih over a broad-voltage range (Supplementary Fig. S1A). At maximum activation (−130 mV), TMEV-reduced Ih amplitudes to 66% in layer 5 neurons (n = 29), compared with such neurons in sham-injected mice (n = 25, P < 0.003, Fig. 1C). Comparing Ih densities to exclude confounding cell-size effects showed a likewise reduction to 77% (P < 0.05, Supplementary Fig. S1B). The current reduction was accompanied by a slightly hyperpolarized voltage dependence of Ih activation (ctrl: V1/2 = –85.8 ± 3.3 mV, TMEV: V1/2 = –90.4 ± 5.5 mV, P < 0.001, Fig. 1D). This shift cannot cause the reduction of maximal Ih, but may increase it at physiological membrane potentials (Fig. 1D, Supplementary Fig. S1A). Thus, type I IFN elevation in a pathophysiological context such as viral CNS infection appears sufficient to inhibit Ih.
Type I IFNs Reduce Neocortical Ih when Applied Acutely and Directly
To study the IFN effect in more detail, we acutely applied recombinant type I IFNs derived from Chinese hamster ovary cells to rat slices containing the somatosensory cortex. The concentration we used (1000 IU mL−1) had an assured effect on neurons (Hadjilambreva et al. 2005) and is in the range observed after systemic viral infections (Heremans et al. 1980). The treatment reduced Ih peak amplitude in all neurons, on average to 80 ± 14% for IFN-α (n = 10, P < 0.05) and to 75.5 ± 19.1% for IFN-β (n = 16, P < 0.001, Fig. 2A) at −130 mV for an example of the entire voltage range (Supplementary Fig. S2A,B).
Acute type I IFN induced amplitude reduction was not due to a change in voltage sensitivity. Both the half-maximum activation, V1/2 and the steepness of the activation curve, k were not affected; neither by IFN-α (ctrl: V1/2 = –88.3 ± 3.4 mV vs. IFN-α: V1/2 = –86.6 ± 4.3 mV, P > 0.16; ctrl: k = 10.6 ± 1.0 vs. IFN-α: k = 10.0 ± 2.1, P > 0.06; n = 10) nor by IFN-β (ctrl: V1/2=–88.3 ± 7.5 mV vs. IFN-β: V1/2 = –89.6 ± 7.9 mV, P > 0.16; ctrl: k = 10.6 ± 2.4 vs. IFN-β: k = 10.8 ± 2.2, P > 0.62; n = 16; Fig. 2B,C). As the effect of both type I IFNs appeared similar, results of further experiments are shown just for IFN-β. The reduction in peak amplitude developed surprisingly rapid for a cytokine, starting at about 10 min after IFN-β arrived at the slice and took about 15 further minutes to reach the maximum (Fig. 2D). The peak amplitude partially recovered when IFN-β was washed out (Supplementary Fig. S2A–C). In the presence of the specific Ih blocker ZD7288 (50 µM), IFN-β did not change the residual non-Ih mediated, hyperpolarization-evoked currents (at –130 mV: 495 ± 130 vs. 502 ± 154 pA, n = 9, P > 0.86; Fig. 2E), further supporting the idea that the current reduction was specific to HCN channels. Consistent with the contribution of Ih to the resting membrane potential, application of IFN-β hyperpolarized layer 5 neurons by 3.3 ± 2.4 mV (n = 9, P < 0.01; Fig. 2F). The Ih reversal potential, determined from the fully reconstructed current-voltage relationships, did not change due to IFN (tested for IFN-β: ctrl: –20.5 ± 1.4 mV vs. IFN-β: –22.1 ± 5.2 mV; n = 16, P > 0.6), and the effective maximum Ih conductance reduced significantly from 13.1 ± 11.3 to 9.9 ± 8.2nS (P < 0.001, n = 16, Supplementary Fig. S2E). On the whole, these data suggest that IFNs rapidly and reversibly decrease the HCN channel conductance, either by reducing channel number or single-channel conductance.
Ih Reduction is due to a Pronounced Inhibition of HCN1 Channels
In addition to the reduction of the amplitude, type I IFNs decelerated Ih kinetics in rat cortical neurons. The Ih activation was best described by double-exponential fits (Fig. 3A, Supplementary Fig. S3A). IFN-β induced a selective increase in the faster time constant by a factor of 1.31 ± 0.42 (n = 16, P < 0.05), whereas the slower time constant remained the same (Fig. 3B). This was further reflected in the reduction of the relative amplitude contribution of the fast component (0.70 ± 0.09 in ctrl vs. 0.61 ± 0.06 under IFN-β; P < 0.01, Fig. 3C). In conjunction with the slowing of the activation, IFNs prolonged the deactivation of Ih by a factor of 1.5 ± 0.65 (n = 16, P < 0.01, Fig. 3D, Supplementary Fig. S3C–E). These effects were qualitatively reproduced by IFN-α (Supplementary Fig. S3C) and partly by TMEV infection (Supplementary Fig. S3H).
Because acutely applied type I IFNs predominantly decreased the fast component of Ih activation, we assumed an ion channel subtype specificity of the effect, namely a reduction of the fastest activating subtype, HCN1. To test this hypothesis, we used a NEURON model based on a reconstruction of a layer 5 neuron's morphology. In the first instance, we fitted the model parameters (Supplementary Experimental Procedures) to our experimental results before IFN application. We then simulated 2 putative mechanisms of HCN channel inhibition by IFNs: 1) a uniform reduction of both HCN1 and HCN2 channel subtype peak conductances, , versus 2) a specific reduction of the HCN1 channel peak conductance, . To attenuate the Ih amplitude to 76%, as in vitro measured at the soma, we reduced for both settings. This required a combined reduction of to 61.2% or a reduction of to 42.5%. The specific HCN1 reduction mimicked our in vitro experimental results (Fig. 4A): The fast time constant of activation decelerated by a factor of 1.2 (ctrl: τfast = 78 ms; IFN-β: τfast = 93 ms; relative amplitude of τfast: 0.61 vs. 0.59), whereas the slow time constant of activation remained almost unchanged (ctrl: τslow = 451 ms; IFN-β: τslow = 491 ms) and the deactivation time constant increased by 1.6 (ctrl: τ = 212 ms; IFN-β: τ = 340 ms). In contrast, modeling a uniform reduction of both HCN subunit models only marginally increased the time constants of activation (τfast = 88 ms, τslow = 482 ms; relative amplitude of τfast: 0.74) corresponding to an increase by the factor of 1.12 and 1.06, respectively, and the time constant of deactivation rather decreased (τ = 210 ms).
To test directly whether the HCN1 subunit is required for the IFN-mediated modulation of Ih, we first isolated and then measured Ih in HCN1−/− mice (Nolan et al. 2003) before and after the application of recombinant mouse IFN-β. In accordance with previous studies (Chen, Shu, Kennedy et al. 2009), Ih in neocortical neurons in these mice amounted to one-third of the one in their HCN1+/+ litters (Fig. 4B,C). In line with our model predictions, IFN-β did not affect the remaining Ih in HCN1−/− mice, presumably mediated by HCN2 channels, whereas in control HCN1+/+ mice we observed a similar attenuation as in rats. In detail, 1000 IU mL−1 IFN-β reduced Ih in HCN1+/+ to 71.9 ± 15.4% (P < 0.01, n = 8), but left it unchanged in HCN1−/− (99.7 ± 19.2%, P > 0.4, n = 8).
If HCN channels were composed of homomers, one would expect a distinct V1/2 shift upon selective HCN1 reduction, in particular when the intracellular cAMP levels are low. To test this assumption, we removed cAMP from the intracellular solution. Also under this condition, IFN-β reduced Ih to 73 ± 16% (P < 0.05, n = 6; Supplementary Fig. S3F) without changing the voltage dependence of channel activation (ctrl: V1/2 = –91.2 ± 4 mV vs. IFN-β: V1/2 = –89.1 ± 3 mV, P > 0.24, n = 6; Supplementary Fig. S3G). When rerunning our model adjusted to the data obtained in vitro, where omitting cAMP shifted the control V1/2 by −2.9 mV without reaching the significance levels (P = 0.14, compared with 100 µM cAMP), reducing the HCN1 conductance to 42.5% led to the prediction of a V1/2 shift by −0.88 mV. This is below our experimental resolution limit.
Taken together, these results are consistent with a specific inhibition of HCN1-mediated Ih by type I IFNs.
IFN-β Inhibits Neuronal Ih without Glial Intermediation
To distinguish a direct neuronal modulation from indirect glial effects, we used primary cultured cortical neurons measured at 9–14 days in vitro. In these cultures, the amount of confounding glial cells was reduced to < 5%. Similar to the observations in slices, application of IFN-β to these cultures led to a reduction of the Ih peak amplitude in all cortical neurons to 75.7 ± 14.9% (n = 5, P < 0.05; Fig. 5A). Consistent with the decreased HCN conductance, the input resistance increased on average 14% in current-clamp recordings (P < 0.001, n = 6; Fig. 5A). These results suggest that the subthreshold effects of IFN-β on cortical excitability are directly attributable to neuronal mechanisms.
IFN-β-Induced Ih Inhibition Requires an Intact Canonical Type I IFN Receptor Pathway
Does IFN-β act directly on HCN channels? To address this question, we expressed the prominent cortical HCN channel subunit HCN1 alone and/or together with HCN2 under conditions where the rat IFN-β signaling pathway is negligible. Given the relative insensitivity of human IFN receptors to rat IFN (Novick et al. 1994) and no confounding currents in the hyperpolarizing range (Supplementary Fig. S5B), we chose the HEK293 expression system. This approach revealed that rat IFN-β did not directly affect currents mediated by rat HCN1 (ctrl: 2.25 ± 1.6 nA vs. IFN-β: 2.20 ± 1.5 nA, n = 6, P = 0.35, Fig. 5B left), by rat HCN1/2 (ctrl: 2.39 ± 1.65 nA vs. IFN-β: 2.40 ± 1.67 nA, n = 9, P > 0.93, Fig. 5B right), and by rat HCN2 (Supplementary Fig. S5A) in cells lacking the specific rat type I IFN receptors. The result points to an IFN-β effect mediated by intracellular signaling cascades rather than a direct conformational change in the channel.
IFN-β actions are mediated by 2 transmembrane receptor chains (IFNAR1 and 2) that form the specific type I IFN membrane receptors (IFNARs; Takaoka and Yanai 2006). First, we investigated whether both IFNAR chains are actually expressed by resident cells in the cerebral cortex. Quantitative real-time PCR from cortical tissue showed expression of IFNAR1 and IFNAR2 mRNA. A detailed analysis in primary cultures indicated that both receptor chains are present in neurons, astrocytes, and microglia (Fig. 6A). Immunohistochemical staining of rat cortex revealed IFNAR1 in neurons (Fig. 6B, for antibody specificity see Supplementary Fig. S6A). IFNAR1 staining on primary cortical neurons showed a punctuated immunoreaction in the cell bodies and the dendrites (Fig. 6C). To determine whether this neuronal expression of IFNAR1 is functional, we tested the activation of signaling molecules downstream of the intracellular domain of the IFNAR subsequent to IFN-β binding. The signaling cascade involves Tyk2 and JAK1, both enzymes of the JAK family, and the signal transducers and activators of transcription (STAT)1 and STAT2 (Takaoka and Yanai 2006). Alternate type I IFN signaling pathways operate via the Ca2+-independent di-acylglycerol-mediated protein kinase C pathway or via the MAP kinase pathway, which include p38 and the extracellular signal-regulated kinase 2 (Takaoka and Yanai 2006). For a selective signal cascade analysis, we incubated primary neocortical neurons for 30 min with bath solution containing 1000 IU mL−1 IFN-β. We first probed the canonical pathway, represented by Tyk2, for phosphorylation. Subsequent to receptor activation by IFN-β application, we observed a marked phosphorylation of Tyk2 (Fig. 6D), which points to a proper activation of the early signaling steps. However, the alternative p38 MAP kinase pathway was not activated in neurons, even after 30 min of IFN-β treatment (Fig. 6E). Taken together, we conclude that the IFNAR and its canonical pathway are present and functionally active in neurons.
Both the existence of neuronal IFNAR and the lack of a direct effect of IFN-β on the HCN channel subunits prompted us to investigate whether IFNAR downstream signaling cascades are required to inhibit Ih by testing neuronal IFN-β modulation in the presence and absence of JAK selective blockers. Inhibition of the first signal transduction step within the JAK/STAT pathway prevented the Ih attenuation by IFN-β, as tested by application of IFN-β after pretreatment with 75 nM JAK1 inhibitor (Thompson et al. 2002). Upon such treatment, the amplitude of Ih remained at 101.6 ± 15.9% (n = 12, P > 0.96; Fig. 6F). In contrast, AG-490, a preferentially type II receptor associated JAK2 inhibitor (Meydan et al. 1996), was not effective in blocking the IFN-β modulation of Ih (Supplementary Fig. S6B). These data show that JAK1/Tyk2, but not JAK2 blockade, prevents the IFN-β effect, supporting the hypothesis that the IFN-β effect on neurons requires the JAK1/Tyk2 receptor-signaling pathway.
IFN-β Shifts Neuronal Resonance by Ih Modulation
To test whether the IFN-β–mediated HCN1 channel reduction leads to functionally relevant changes in cellular excitability, we explored the impact of IFN-β on resonance. One function of Ih in pyramidal neurons is to complement passive cell properties in generating a membrane resonance leading to a distinct frequency preference of incoming inputs at rest (Ulrich 2002; Narayanan and Johnston 2008).
To examine whether the magnitude of changes in Ih after rising levels of IFN-β is sufficient to change the resonance behavior, we employed a 3-step approach, starting with an established mathematical model of neuronal resonance (Hutcheon et al. 1996), followed by the application of a morphologically realistic model of a layer 5 neuron, and finalized by experimental testing of the achieved prediction. For the mathematical model, we adjusted parameter values to our experimental data (Supplementary Experimental Procedures). We modeled the effect of IFN-β according to the observed alterations: We 1) reduced to 9.9nS, 2) increased τfast to 0.312 s, and 3) decreased pf to 0.61. This resulted in a resonance frequency shift from 2.23 to 1.76 Hz and an increase in peak impedance (|Z|) from 121.3 to 131.2 MΩ, accompanied by a shift in Φ0 (the frequency where the phase shift plot passes zero) from 1.55 to 1.16 Hz. The subsequently used morphological realistic model of the layer 5 neuron (Supplementary Experimental Procedures) showed similar results when HCN1 was specifically reduced (Fig. 7A,B). Here, the resonance frequency of the membrane shifted from f = 4.52 to 3.3 Hz, the peak impedance increased from |Z| = 36 to 40 MΩ, and Φ0 decreased from Φ0 = 3.0 to 1.95 Hz.
Finally, we tested the prediction of the above models in layer 5 pyramidal neurons of acute brain slices. As predicted by our modeling, IFN-β application shifted the membrane resonance to lower frequencies (ctrl: f = 2.3 ± 0.7 Hz, IFN-β: f = 1.9 ± 0.7 Hz, P < 0.001, n = 9) and increased the maximal membrane impedance (ctrl: |Z| = 117 ± 62 MΩ, IFN-β: |Z| =133 ± 70 MΩ, P < 0.05, n = 9; Fig. 7C,D). In further accordance with the model, we observed an apparent reduction of the inductive phase component along with a phase shift of the input impedance (ctrl: Φ0 = 1.3 ± 0.5 Hz, IFN-β: Φ0 = 1.1 ± 0.5 Hz, P < 0.05, n = 7; Fig. 7E). The shift of the resonance parameters to lower values was associated with a narrower range of preferred intrinsic frequencies (half-band width [HBW]ctrl = 3.9 ± 1.4 Hz, HBWIFN-β = 3.2 ± 0.9 Hz, n = 9, P < 0.05; Fig. 7F). The strength of the resonance (Q-value), represented by the ratio of the impedance at the resonant peak to the impedance at rest, remained unaffected (Qctrl = 1.3 ± 0.24 vs. QIFN-β = 1.3 ± 0.15, n = 9, P > 0.32). On the single cell level, IFN-β led to a preference of lower frequencies.
IFN-β Slows the Cortical EEG
Given the impact of IFN-β on resonance behavior in single cerebral pyramidal neurons in our in vitro slice preparation and the role of resonance in setting network activity, we hypothesized that IFN-β affects the oscillatory dynamics of neuronal networks, which appear at the cerebral cortex as EEG rhythms (Karameh et al. 2006). To test this assumption, we applied an effectual amount of IFN-β directly to the rat cerebral cortex in vivo (Fig. 8A, Supplementary Fig. S8A). A deposit volume of 5 µL containing 5000 IU recombinant rat IFN-β was placed right above the cortex while recording the surface EEG in awake motorically silent rats. The diffusion-driven exogenous IFN-β application significantly slowed the cortical EEG activity by ∼0.6 Hz in the frequency range between 2 and 6 Hz (Fig. 8B–D). Furthermore, the slowing was associated with a decrease in the power decline from 2.44 ± 0.16 to 1.98 ± 0.18 Hz (n = 6, P < 0.05). The EEG alterations were not due to the experimental methodology, as normal saline without IFN-β had no effect on the EEG power (fctrl = 2.19 ± 0.29 Hz, fsaline = 2.21 ± 0.19 Hz, P = 0.72, n = 4; Supplementary Fig. S8B–D).
We repeated cortical EEGs in older mice litters (P32–40) from HCN1+/− matings. The effects of exogenous IFN-β application to cortices of HCN1+/+ mice resembled the ones in rats. Here, IFN-β reduced the power of the higher frequencies in the cortical EEG activity, and this was associated with a decrease in the power decline from 2.23 ± 0.62 to 1.40 ± 0.25 Hz (n = 6, P < 0.01; Fig. 9B–D, upper row) comparable with the one found in the rats. When applied to cortices of litter HCN1−/− mice, however, IFN-β did not alter the EEG (power decline: 1.09 ± 0.20 vs. 1.11 ± 0.23 Hz, n = 7, P = 0.49; Fig. 9B–D, lower row).
These findings suggest that IFN-β modulates spontaneous EEG slow-wave activity depending on the presence of HCN1 and can reversibly alter the physiological responses of cortical neuronal networks.
The present study demonstrates that Ih, a major component of intrinsic neuronal excitability, is instrumental in mediating type I IFN-induced changes in neuronal excitability. Extended type I IFN presence as after in vivo induction of IFNs in the CNS by Theiler's virus infection led to an Ih reduction. Also, the acute application of type I IFNs rapidly reduces rodent Ih by about one-fourth. The effect is attributable to a modification of HCN channels, because, first, after blocking Ih no change in current amplitude upon IFN application could be observed, and secondly, synaptic influences or other voltage sensitive currents were excluded by pharmacologically isolating Ih. Based on modeling results and measurements in HCN1−/− mice, we conclude that the effect is predominantly mediated by the fast activating HCN subunit, HCN1. Data from primary cultures and the detection of functionally intact IFNAR in neurons revealed that IFN-β directly acts on neurons, that is, without glial mediation.
Interactions of the Signaling Cascades of IFNAR with HCN Channels
Classical signaling in response to IFNAR activation includes a number of phosphorylation steps. Here, we show that cortical neurons possess an essential component of the IFN-signaling cascade, functional IFNAR. Furthermore, the effect of IFN-β on HCN channels requires the activation of the JAK/STAT pathway since IFN-β did not interact directly with HCN channels and was ineffective after disrupting the signaling cascade.
However, the connection between IFNAR activation and HCN channels remains to be demonstrated. One possible link is the p38 MAP kinase, as it can be activated by IFNARs (Takaoka and Yanai 2006) and exerts a direct influence on Ih (Poolos et al. 2006). Nevertheless, 2 lines of evidence in our experiments argue against such an interaction: First, IFN-β failed to phosphorylate neuronal p38 MAP kinase in cortical neurons. This also excludes a contribution of MAP kinase-induced arachidonic acid to be involved in the decrease of maximal current (Fogle et al. 2007). Secondly, IFN-β does not cause a hyperpolarization in the voltage dependence of Ih activation, whereas an upregulation of p38 MAP kinase resulted in an ∼11-mV depolarization of in hippocampal pyramidal neurons (Poolos et al. 2006). Due to the lack of a shift of V1/2, we also exclude an IFN-β/HCN channel interaction through allosteric regulators of voltage dependence of activation such as cAMP, H+, 4,5-PIP2, or signaling lipids (Fogle et al. 2007). A sole protein–protein interaction (i.e. exclusive phosphorylation) appears also unlikely because of the somewhat delayed onset of the effect. Further, the changes of neuronal properties such as input resistance and membrane time constant, which may be linked to HCN channels, were dependent on protein synthesis (Beyer et al. 2009). Therefore, future research on the exact molecular mechanism may be focused on proteins involved in both the IFNAR signaling pathway and HCN modulation and newly transcribed upon IFNAR activation.
The IFN-β Modulation of Ih is Sufficient to Cause Alterations in Functional Properties of Cortical Pyramidal Neurons
Ih is involved in the regulation of neuronal excitability particularly due to its partially open state at resting membrane potential (Wahl-Schott and Biel 2009). Accordingly, the input resistance was markedly increased by IFN-β in neocortical neurons in brain slices (Hadjilambreva et al. 2005) and in primary cultured cortical neurons (this study). Given the distance dependence of the HCN channel density in dendrites of pyramidal neurons (Magee 1998; Kole et al. 2006), an inhibition of HCN channels might lead to an even greater augmentation in dendritic excitability by decreasing the local resting conductance and increasing the summation of excitatory postsynaptic potentials (Magee 1998; Huang et al. 2009). Their local action in dendrites remains to be tested. Of importance for dendritic integration and neuronal computation is the frequency-dependent response of neurons at and below resting membrane potential, for which Ih is the main responsible active current (Ulrich 2002; Narayanan and Johnston 2008). Eventually, resonance describes the frequency dependence of impedance and the band-pass filter characteristics of dendrites for incoming signals (Ulrich 2002). Furthermore, the accompanying phase shift endows neurons to scale the arrival times of signals on the soma (Narayanan and Johnston 2008). Our data suggest that layer 5 neurons favor responses to lower frequency inputs when exposed to elevations in IFN-β. Taken together, these changes point to a significant modulation of neuronal excitability by IFN-β–induced Ih reduction.
Connection between Ih Reduction and EEG Slowing
A number of points suggest that Ih modulation considerably contributed to observed changes in the EEG. As strongest argument, we regard the lack of an IFN-β effect on the cortical EEG of HCN1−/− mice (this study). Further, HCN channels exert a global control of neuronal network rhythms (Wahl-Schott and Biel 2009), and the appearance of aberrant EEG activity and seizures is associated with HCN channel reduction (Strauss et al. 2004; Kole et al. 2007; Marcelin et al. 2009). We observed a reduced EEG power in the frequency range between 1 and 8 Hz associated with IFN-β, as it was also reported for Ih inhibition in frontal lobe epilepsy (Marcelin et al. 2009). Karameh et al. (2006) suggested with a modeling approach that cortical alpha rhythms depend on intrinsic currents in layer 5 cells, namely Ih and T-type calcium current. Their study also predicts a pronounced shift to the delta range upon Ih blockade. Furthermore, the sensitivity to synchronized synaptic inputs is promoted by hyperpolarization (Carr et al. 2007) as it is triggered by IFN-β (this study). Interestingly, EEG changes have long been recognized after treatment with IFN-α (Dafny 1983; Birmanns et al. 1990; Kamei et al. 2005), the other type I IFN that also activates signaling via IFNAR.
In the process of inflammation, IFN-β levels are dynamic and dependent on the local environment. Therefore, the concentration used in this study might only represent one within the range of local IFN-β concentrations triggered by inflammation. At present, the precise extracellular type I IFN protein levels that might be expected during a CNS inflammation are unknown, although there are some hints. For example, neuronal IFN-β production in viral infection has been investigated in mice by RT-PCR providing quantitative values of IFN-mRNA production (Delhaye et al. 2006), and a time line of nonspecified IFN tissue content was studied by a 3H-uridine-based assay (Heremans et al. 1980). Quantitative protein analysis of spinal cord homogenate during experimental autoimmune encephalomyelitis, an animal model of multiple sclerosis, showed that IFN-β is expressed at markedly higher amounts in the CNS than in the periphery (Prinz et al. 2008). Given that in inflammation cytokines generally act together, they may combine their impact in vivo, in particular concerning the type I IFNs acting at the IFNAR. Due to the local production (Delhaye et al. 2006), IFNs may activate their respective receptors, even if the tissue concentrations are still quite low. By utilizing the CNS inflammation caused by the neurovirulent GDVII strain of TMEV at the height of type I IFN production (Delhaye et al. 2006), we here showed that viral IFN induction mimic the amplitude reduction in Ih. This supports a pathophysiological role of the type I IFN neuron interaction in CNS inflammation.
Phenomenological similarities between Ih reduction and inflammation provide indirect evidence for a link between both. Pharmacological HCN1 reduction with compounds such as ketamine or several volatile anesthetics (Chen, Shu, Bayliss et al. 2009; Chen, Shu, Kennedy et al. 2009) produces an inflammation-like deteriorated mental state, including disturbed vigilance. Likewise, subtle modulation of HCN channel activity, as observed with physiological changes of cAMP levels, contributes substantially to altered network activity correlated with behavioral states (Wahl-Schott and Biel 2009). In human inflammation states, such functional changes may associate with sickness behavior and an increased susceptibility to depression (Amodio et al. 2005). It was suggested that certain cytokines play a causal role in the genesis of psychosocial alterations (Kent et al. 1992; Dantzer et al. 2008) and that IFN-α therapy can produce such side effects and may contribute to cytokine-related subtypes of affective disorders (Anisman et al. 2008; Dantzer et al. 2008).
Ih modulation by IFN-β may even be of broader importance under physiological conditions given the basic IFN-β level recently reported in the uninfected and noninflamed CNS (Prinz et al. 2008). This would imply a physiological neuromodulatory role of IFN-β.
In summary, our data imply a major role of IFNs in altering the neuronal state during inflammation. This put IFNs in line with previously recognized neuromodulatory cytokines such as IL-2, IL-1β, IL-6, and TNF-α (Mendoza-Fernandez et al. 2000; Dantzer et al. 2008). Taking the contribution of Ih for the proper function of neuronal cells in the nervous system (central and peripheral) into account, our experimental data open up new fields of investigation. These results might shed light on the involvement of Ih alterations in the adaptive processes during acute and chronic neuronal inflammation (Kent et al. 1992; Johnson 2002). Of particular interest will be to determine whether IFNs counteract the frequency response tuning capabilities of neurons, or if they act as a natural protector of brain tissue from inflammation (Prinz et al. 2008) by providing an auxiliary mechanism to adapt neuronal computation to the state of inflammation.
This study, in particular the work of K.S. and U.S., was supported by the German Research Foundation (DFG STR865/3-1) and the DAAD/Go8 program (U.S. and M.H.P.K.). Some of the equipment used was donated by the Sonnenfeld-Stiftung, Berlin (A.U.B. and U.S.).
We thank Bettina Brokowski, Carla Strauss, and Rike Dannenberg for expert technical assistance, Shigetada Nakanishi for the donation of rHCN2, and Thomas Michiels for advice on the induction of the viral encephalomyelitis and for providing TMEV GDVII. We also thank Arndt Rolfs for providing us with laboratory space, consumables, and equipment in the initial phase. Conflict of Interest: None declared.